OBJECTIVES

: To develop a novel and simple method for accurate mapping of quality-of-life (QOL) scores to utility values without under-estimation of variance or dilution of association.

METHODS

: We propose a novel method called Mean Rank Method. It is similar to the Equipercentile method but it does not require smoothing of the cumulative distribution functions. Two series of data values are mapped according to ranks (if no ties) or mean of ranks (if there are ties). We performed simulations and applied the method to two datasets from Singapore to map a generic and a cancer-specific quality-of-life (QOL) scale to the EQ-5D-5L utility index to empirically assess the method's performance and compare it against the ordinary least squares (OLS) and Equipercentile methods.

RESULTS

: In a range of realistic scenarios simulated, the Mean Rank, Equipercentile and OLS methods under-estimated the variance of the utility values by about 0.5%, 2% and 50%, respectively. The Mean Rank Method over-estimated the regression coefficient between the true utility and a covariate by about 2%, as compared to about 5% and 25% under-estimation by the Equipercentile and OLS methods, respectively. Findings about variance shrinkage and relationship with clinical covariates obtained from the two datasets were similar to the simulation results. However, at the individual level of prediction, none of the three methods consistently out-performed the others in terms of mean squared error, mean absolute error or intraclass correlation coefficient.

CONCLUSIONS

: Utility values obtained from the Mean Rank Method accurately shows the distribution features of the observed utilities at the population and group levels. It is advantageous over the Equipercentile method in terms of not requiring complicated smoothing procedures.